Agentic Fleet

A fortified enterprise agent platform — three institutional agents on Google Cloud, discoverable, accountable, and safe.

Comment

Agentic Fleet — Fortified Enterprise Agent Platform

Agentic Fleet is a production-ready, enterprise-grade multi-agent orchestration platform. It deploys a secure, scalable network of specialized institutional agents directly integrated with cloud infrastructure—fulfilling the rigorous lifecycle, security, and governance demands of modern enterprise environments.

Key Capabilities & Architecture

  • Central Agent Registry — Publish, version, and discover verified enterprise agents with tool authorization and granular metadata. Includes three live pre-configured agents:

    • Finance / AP: Handles invoice workflows and financial data extraction.
    • Compliance: Evaluates regulatory checks and policy verdicts.
    • Ops-Monitoring: Tracks infrastructure telemetry and operational health.

  • Decoupled Asynchronous Runtime — Engineered for long-running workflows using Google Cloud Pub/Sub and an isolated Cloud Run worker. Submit tasks asynchronously, track execution via unique job_ids, and poll for results.

  • Cross-Agent Collaboration & Handoff — Enables dynamic delegation between specialized agents (e.g., Finance/AP discovers and hands off an invoice to Compliance for policy approval).

  • Zero-Trust Agent Identity & Gateway — Unified gateway routing with strict policy enforcement. Invocations require principal attribution, and only active, registry-published agents can be executed.

  • Model Armor (Multi-Layer Inline Guardrails) — Built-in defense-in-depth security featuring prompt-injection detection, automated PII redaction, and native Gemini safety filters. Every redaction and blocked request is recorded in the audit trail.

  • Persistent Memory Bank — Secure, cross-session contextual memory powered by Google Cloud SQL (PostgreSQL).

  • End-to-End Observability & Governance — Immutable audit logging capturing every routing decision, safety intervention, and agent transaction, paired with a live read-only dashboard.

Technologies Used

  • LLM Engine: Gemini 3.5 Flash (via the Gemini API & google-genai SDK, model env-swappable)
  • Compute: Google Cloud Run (main API gateway service + async worker)
  • Messaging & Queue: Google Cloud Pub/Sub (async job queue)
  • Database & Persistence: Google Cloud SQL (PostgreSQL — registry, memory, audit trail, jobs)
  • Backend Framework: Python 3.11, Flask, Gunicorn
  • CI/CD & Containers: Google Cloud Build, Artifact Registry

Links

Built With

Share this project:

Updates